Are regionally calibrated earthquake forecast models more informative than global models? First results from California and Italy

José A. Bayona, William H. Savran, & Maximilian J. Werner

Submitted August 16, 2021, SCEC Contribution #11537, 2021 SCEC Annual Meeting Poster #264

Earthquake forecasting models formalize critical hypotheses about seismogenesis that can support regional and global seismic hazard assessments. Regional models provide detailed seismicity forecasts due to the increasing availability of high-resolution geophysical datasets, while global models offer greater testability given the higher seismic activity observed worldwide. Until now, comparisons between global and regional earthquake models have largely remained qualitative, thereby lacking quantitative information to formally contrast their strengths and weaknesses. Here, we prospectively evaluate the predictive skills of global and regional seismicity forecasts in California and Italy using a set of consistency and comparative likelihood tests implemented in the Collaboratory for the Study of Earthquake Predictability (CSEP). To this end, we project the Global Earthquake Rate Activity (GEAR1) model onto the two CSEP testing regions, and extrapolate its rates to a magnitude threshold of 4.95 assuming a b-value of unity. In addition, we use 32 Mw ≥ 4.95 earthquakes reported in California during January 1, 2014 and December 31, 2020, such as the 2019 Mw = 7.1 Ridgecrest earthquake, and 12 target events in Italy, including the 2016 Mw = 6.6 Marche-Umbria earthquake. Prospective test results show that, firstly, in California, all models overpredict the number of earthquakes and fail to accurately forecast the spatial distribution of seismicity. Further, a model that adaptively smooths regional epicentral locations and a neokinematic model constructed from tectonic and geodetic data obtain information gain scores per earthquake (IGPE) of 1.0 and -0.7 over GEAR1, respectively. Secondly, in Italy, all models forecast rates consistent with the observations and, with the exception of models depending on spatial variations of the b-value, models can satisfactorily explain the spatial distribution of epicenters. Moreover, an adaptively smoothed seismicity model and a model based on fault slip rates obtain IGPEs of 0.8 and 0.7 over GEAR1, respectively. Thus, we find that regionally calibrated smoothed seismicity models can forecast seismicity in California and Italy significantly better than the global GEAR1 model. Currently, we conduct similar analyses in other CSEP testing regions to provide a more comprehensive description of regional and global model performance and to explore whether the tectonic setting plays a key role in the predictability of earthquakes.

Key Words
Earthquake forecasting, statistical seismology

Bayona, J. A., Savran, W. H., & Werner, M. J. (2021, 08). Are regionally calibrated earthquake forecast models more informative than global models? First results from California and Italy. Poster Presentation at 2021 SCEC Annual Meeting.

Related Projects & Working Groups
Earthquake Forecasting and Predictability (EFP)